Micro-vision Image Stitching System for Large-scale and Fine-featured Circuit Substrates
Yuanyang Wei, Jian Gao, Yongbin Zhong, Lanyu Zhang · 2021
With the improvement of the manufacturing process and packaging technology of the precision electronic manufacturing industry, various precision circuits with large scale and fine features are widely used in high-end precision electronic products and instruments. At present, large-scale circuit substrates can reach very fine pitch and line width in only several micron-meter level. So the quality inspection equipment should have the inspection capabilities of large-scale and high-precision detection. Since a single view of image with high-resolution only have a small field of view, image stitching by registration and fusion is an effective method to inspect the whole large scale substrate. Hence, this paper proposes a micro vision image stitching method with high precision through coarse and fine registration. The real-time position information of the platform is used to calculate the approximate pixel offset between images to achieve coarse registration. Then, two convolutional neural regression networks connected by a spatial transformation network are used to achieve fine registration. A high-precision nano-positioning platform is used to establish image registration dataset, which is used for the training of convolutional neural regression networks. Experimental results show that under the condition of resolution of 0.5um//pixel and stitching overlap of 20%, our method can achieve less than 0.2 pixel registration error and less than 25ms registration time for a single image.